Continuous Power Management of Decentralized DC Microgrid Based on Transitional Operation Modes under System Uncertainty and Sensor Failure

Abstract

Continuous power management for a decentralized DC microgrid (DCMG) is proposed in this study to achieve power balance and voltage regulation even under system uncertainty and voltage sensor failure. The DCMG system achieves continuous power management through only the primary controller to reduce the computational burden of each power agent. To enhance the reliability and resilience of the DCMG system under DC bus voltage (DCV) sensor failure, a DCV sensor fault detection algorithm is suggested. In this algorithm, DCV sensor failure is detected by comparing the measured DCV with the estimated DCV. If power agents identify the failure of the DCV sensor, it changes the operation properly according to the proposed control mode decision algorithm to guarantee the stability of the DCMG system. When uncertain conditions like sudden grid disconnection, DCV sensor failure, electricity price change, power variation in distributed generations, and critical battery status occur, the DCMG system is changed to transitional operation modes. These transitional operation modes are employed to transmit the power agent information to other agents without digital communication links (DCLs) and to accomplish power sharing even under such uncertain conditions. In the transitional operation modes of the DCMG system, the DCV levels are temporarily shifted to an appropriate level, enabling each power agent to detect the uncertainty conditions, and subsequently to determine its operation modes based on the DCV levels. The reliability and effectiveness of the proposed control strategy are confirmed via various simulation and experimental tests under different operating conditions.

1. Introduction

Interest in sustainable and renewable energy is increasing due to environmental problems such as greenhouse gas emissions and fossil fuel depletion [1,2]. With the increasing demand for cleaner energy, the microgrid system has emerged as a promising solution to generate and distribute electrical energy. By using the microgrid, the distributed generations based on renewable energy sources, utility grid, energy storage systems (ESSs), and loads are interconnected efficiently and reliably to overcome the limitation caused by the intermittent nature of renewable energy, which provides a sustainable alternative to the conventional power grid [3,4,5]. By harnessing diverse renewable energy sources, microgrids contribute to reducing the carbon footprint and enhance energy resilience. Microgrid systems that are efficient and robust for the use of distributed energy sources are usually composed of ESSs, loads, and distributed generations such as a wind turbine and photovoltaic (PV) device [6].

Depending on the type of power supplied, the microgrid configuration is divided into the DC microgrid, AC microgrid (ACMG), and hybrid AC/DC microgrid [7]. Among these structures, DCMG stands out for its simplicity and efficiency. Unlike hybrid AC/DC and AC microgrids, the DCMG minimizes unnecessary power conversion stages and reduces technical challenges such as synchronization, harmonics, and reactive power [8,9,10,11]. Furthermore, DC energy sources such as fuel cells, PV, supercapacitors, or batteries are easily interconnected in the DCMG system [12].

According to the configuration of the digital communication link (DCL), the control strategy for the DCMG system is categorized into three main types: centralized control, decentralized control, and distributed control [13,14,15]. The centralized control strategy uses a central controller that communicates with all power agents via DCLs to determine the operation modes of each power agent [16,17]. The distributed control strategy also uses DCL to communicate with adjacent power agents. However, in the distributed control method, power agents determine the proper operation modes with a limited DCL to adjacent power agents without relying on a central controller [18,19]. It is important to note that the transmission time delay caused by the DCL that is unavoidable in both the distributed and centralized controls might negatively impact the efficiency and stability of the DCMG system [3]. A decentralized control strategy can be employed as an alternative way to solve this weakness. In the decentralized control method, the operation modes of the DCMG system are properly determined based on only local measurements such as the DCV, without using any DCL between power agents [20]. In addition, power agents normally operate independently in the decentralized DCMG system without the use of any DCL, which significantly increases the simplicity, scalability, and flexibility of the DCMG system [21,22].

Generally, the DCV is utilized as global information for powers agent in the decentralized DCMG system [23,24,25,26]. If the DCV decreases from the nominal DCV value, it indicates that the power supply to the DC bus is less than the power demand. In contrast, if the DCV increases from the nominal DCV value, it means that there is power surplus in the DCMG system. Traditionally, the droop control strategy is employed in decentralized DCMG systems to control the DCV to the nominal value [27,28,29,30,31]. In the study in [29], DCMG power balance is achieved by using a voltage–current (V-I) droop controller according to the DCV level. To overcome the weakness of the DCV fluctuation in this scheme, the researchers in [31,32] proposed the use of an integral term in the secondary controller for ensuring DCV restoration to the nominal value. The secondary control strategy is used in these methods to regulate the DCV to its nominal value, while the primary control strategy is employed to maintain the power balance in the decentralized DCMG system. Although the voltage fluctuation problem caused by the droop controller is mitigated by combining the primary and secondary controls, there still exists a computational burden in the above methods. Moreover, it is desirable to consider a method to enhance system reliability even under sensor device failure.

To operate the DC microgrid efficiently, several sensors are utilized to provide feedback signals for real-time monitoring and control systems. It is worth mentioning that the information from these sensors plays a significant role, especially in decentralized DCMG configuration, to determine the operation modes of the power agent because of the absence of the DCLs between the power agents. The sensor failure of abnormality leads to discrepancies in data interpretation among power agents, potentially resulting in suboptimal system operation, or even instability in an extreme case [33,34]. To overcome such sensor failure, the researchers in the studies in [35,36] presented fault-tolerant control via hardware redundancy. In those methods, additional sensors were equipped as redundancy hardware to deal with sensor faults. In spite of the reliability enhancement against sensor failures, these approaches obviously increase system costs. Fault detection and isolation are also proposed for the distributed DCMG system via an adaptive observer method to detect different types of sensor failure [37]. The research in [38] proposed an H_∞ observer for DC–DC converters in order to compensate for current and voltage sensor faults. However, it is difficult to implement those control strategies for the decentralized DCMG system because the DCLs between power agents are absent in the decentralized DCMG.

Motivated by this concern, this study presents a continuous power management strategy of a decentralized DCMG system based on the transitional operation modes under system uncertainty and sensor failure to improve system reliability. To achieve both a power balance and voltage stabilization continuously under various uncertain conditions such as power variation in distributed generation, grid disconnection, and critical battery status, the transitional operation modes are employed to transmit the power agent information to others without DCLs. In these transitional operation modes of the DCMG system, the DCV levels are temporarily adjusted to a proper level. Furthermore, to minimize the electricity price, the proposed control strategy determines the operation modes of power agents by considering the electricity price conditions in the grid-connected mode (GCM). It is important to mention that the proposed control strategy uses only primary control to achieve DCV restoration, which significantly reduces the computational burden.

To monitor the abnormality of DCV sensors, all the power agents estimate the DCV by using the observer in the proposed scheme. For this aim, the proposed DCV sensor fault detection algorithm is implemented in all power agents to detect the DCV sensor fault by comparing the estimated DCV with the measured DCV. Under normal conditions without DCV sensor failure, the proposed power management strategy achieves both power balance and voltage regulation by using the estimated DCV. When the DCV sensor failure occurs in any power agent, the operation modes of the power agent are changed properly based on the proposed control mode decision algorithms to ensure a continuous power management strategy even in the presence of DCV sensor failure. The main contributions of this study are summarized as follows:

(i)

The proposed control strategy based on the decentralized DCMG configuration achieves continuous power management based on the transitional operation modes under system uncertainty and sensor failure, which greatly improves the system reliability. In addition, both voltage regulation and power sharing are accomplished by using only the primary controller to reduce the computational burden.

(ii)

Even in the presence of uncertainties such as grid disconnection, electricity price changes, power variation in the distributed generation, and critical battery status, the proposed decentralized DCMG system ensures voltage stabilization and power balance by using the transitional operation modes without DCLs. Once an uncertain condition occurs, the DCV levels are temporarily adjusted to a proper level in transitional operation modes, and then, the power agents determine their operation modes based on the DCV levels.

(iii)

The DCV sensor fault detection algorithm continuously monitors the abnormality of the DCV sensor in the proposed scheme to reinforce the DCMG system’s reliability. Under normal conditions without DCV sensor failure, the proposed scheme achieves both voltage regulation and power balance. If DCV sensor failure happens in a power agent, the proposed control mode decision algorithm properly changes the power agent operation into current control mode for seamless power management. In addition, the proposed scheme can stably work even under multiple DCV sensor faults in more than one power agent at the same time if there exists a normal DCV sensor in a DCMG system.

This paper is organized as follows: In Section 2, the DCMG system configuration and decentralized control strategy are described. Section 3 explains the control strategy of power agents in normal conditions, the control strategy under DCV sensor fault, and transitional operation modes of power agents. To validate the proposed control strategies, Section 4 and Section 5 show the test results of the decentralized DCMG system under various conditions by using simulation and experiments. Finally, Section 6 provides the conclusion of this study.

2. System Configuration of a DCMG and Decentralized Control Strategy

A system configuration of a decentralized DCMG is shown in Figure 1. A decentralized DCMG is composed of four power agents: a grid agent, wind turbine agent, load agent, and battery agent. The grid power agent connects the utility grid to the DC bus through a transformer and bidirectional AC–DC converter. The wind turbine power agent connects the wind turbine to the DC bus, in which the mechanical energy is converted into electrical energy via a three-phase permanent magnet synchronous generator (PMSG) and unidirectional AC–DC converter. The battery power agent connects the battery to the DC bus via the bidirectional DC–DC converter. The load agent consists of DC loads directly connected to the DC bus.

In this DCMG system, the battery and grid power agents are capable of transferring bidirectional power, by supplying the power into the DC bus, or absorbing the power from the DC bus. On the other hand, the wind turbine agent only has unidirectional power flow, supplying the power to the DC bus. As shown in Figure 1, $P_L$, $P_W$, $P_G$, and $P_B$ represent the power flow of the load, wind turbine, grid, and battery agents. In this paper, positive power values ($+$) represent when the power agent absorbs the power from the DC bus, while negative power values ($-$) represent when the power agent supplies the power to the DC bus.

Figure 2 illustrates the simplified structure of the power converter for agent i, which consists of a switch module; an inductor, $L_i$; a capacitor, $C_i$; virtual resistance, $R_{Li}$, for the grid; a wind turbine; and battery agents. In Figure 2, i denotes each agent, i.e., $i$ = G, B, and W, representing the grid, battery, and wind turbine agents, respectively. From Figure 2, the power converter model is derived as follows:

$$\frac{d{V}_{DC,i}}{dt}=\frac{I_i^{out}}{C_i}-\frac{V_{DC,i}}{R_{Li}{C}_i}$$

(1)

where $V_{DC,i}$ and $I_i^{out}$ denote the DCV measured from the sensor and converter output current of agent $i$, respectively. It should be noted that the resistance, $R_{Li}$, for each power agent, $i$, is connected in parallel to produce the total equivalent resistance, $R_L$, at the DC bus of the DCMG system. This yields the following relation:

$$\frac{1}{R_L}=\sum \frac{1}{R_{Li}}$$

(2)

where $R_L$ is the total equivalent load.

3. Control Strategy and Transition Operations of Power Agents under Emergency Conditions

3.1. Control Strategy of Power Agents

Figure 3 represents the control block diagram of agent $i$ in the decentralized DCMG system. In this control block diagram, $V_i^{ref}$ denotes the converter reference voltage and $d_i$ denotes the converter duty ratio of power agent i. Unlike the conventional droop controllers [30,31], the proposed control strategy only utilizes the primary controller that consists of voltage and current controls to minimize computational burden. The primary controller consists of two cascaded PI controllers for voltage control and current control, as shown in Figure 3. This cascaded structure is quite common and typical in power electronic converters such as the three-phase AC–DC converter or three-phase inverter. Also, to ensure power balance even in emergency conditions, the DCV sensor fault detection algorithm is implemented in agent $i$, which is described in detail in Section 3.2. For this purpose, the DCV of agent $i$ is estimated by the observer. The estimated DCV is used to detect DCV sensor fault for the purpose of guaranteeing power balance by continuously monitoring DCV sensor failures with the observer even when the DCV sensor operates normally. In addition, by using the estimated DCV value, agent $i$ regulates the DCV using the proportional–integral (PI) controllers. By applying the conventional observer to (1) with the output error, the DCV of agent i can be estimated as follows:

$${\dot{\widehat{V}}}_{DC,i}=\frac{I_i^{out}}{C_i}-\frac{{\widehat{V}}_{DC,i}}{C_i{R}_{Li}}+N_i(V_{DC,i}-{\widehat{V}}_{DC,i})$$

(3)

where ${\widehat{V}}_{DC,i}$ is the estimated DCV and $N_i$ represents observer gain. To implement the proposed control strategy in both the simulations and experiments, the sampling time was selected as 100 μsec for all power agents to achieve current control. Considering that the current mostly has the fastest dynamics in power converters, and that the sampling time is sufficiently faster than the current dynamics, the influence of the digital control loop delays on voltage regulation and power sharing performance is not as serious as that in the studies in [32,35,38].

To ensure power balance, the decentralized DCMG controller in this study uses nine steady-state operation modes, as shown in

Table 1. Steady-state operation modes.

Mode	Battery Agent	Wind Turbine Agent	Grid Agent	Load Agent
NO1	BCCMdis	MPPT	GVCMinv	Normal
NO2	BCCMdis	MPPT	GVCMcon	Normal
NO3	BCCMchar	MPPT	GVCMinv	Normal
NO4	BCCMchar	MPPT	GVCMcon	Normal
NO5	IDLE	MPPT	GVCMcon	Normal
NO6	IDLE	MPPT	GVCMinv	Normal
NO7	BVCM	MPPT	Fault	Normal
NO8	BCCMchar	VCM	Fault	Normal
NO9	IDLE	VCM	Fault	Normal

, under normal conditions without any DCV sensor fault. In each operation, the control modes of power agent $i$ are determined according to the criterion that only one power agent regulates the DCV at each instant. When the DCMG system is subject to a change of operation modes caused by uncertainties such as power variation, the transition operation modes are used to identify the power agent that regulates the DCV. The transition operation modes are described in detail in Section 3.3. The control modes of power agent $i$ are separated into voltage control and current control. The voltage control mode implemented as shown in Figure 3 is used to regulate the DCV. In contrast, the current control mode is realized only by the PI current controller.

Nine steady-state operation modes in Table 1 can be explained in detail. In NO₁~NO₉, the load agent operates in the normal mode, which indicates that the load agent is connected to the DCMG system without load shedding [31,32,39]. In this paper, the electricity price condition is determined via the real-time pricing (RTP) method as an electricity pricing policy [30,31]. While the steady-state operation modes NO₁~NO₆ represent the grid-connected modes, NO₇~NO₉ represent the islanded modes when the grid agent is disconnected to the DCMG. The power balance is ensured effectively in both the grid-connected and islanded modes.

Operation NO₁: This mode occurs with a high electricity price when the sum of the maximum discharging battery power and wind power is larger than the load. In this case, the grid agent regulates the DCV at a nominal value, V_nom, via GVCM_inv (grid agent voltage control mode by inverter operation). The battery agent operates in BCCM_dis (battery current control mode by discharging operation), and the wind turbine agent works in maximum power point tracking (MPPT) mode.

Operation NO₂: This occurs with a high electricity price when the sum of the maximum discharging battery power and wind power is lower than the load. To guarantee power balance, the grid agent supplies power to the DC bus even with a high electricity price by regulating the DCV at V_nom. The grid agent works in GVCM_con (grid agent voltage control mode by converter operation), the battery agent operates in BCCM_dis, and the wind turbine agent works in the MPPT operation.

Operation NO₃: This occurs with a normal electricity price when the sum of the maximum charging battery power and load is less than the wind power. To ensure power balance, the grid agent absorbs power from the DC bus by regulating the DCV at V_nom. The grid agent works in GVCM_inv, and the battery agent operates in BCCM_char (battery current control mode by charging operation). The wind turbine agent works in the MPPT operation.

Operation NO₄: This occurs with a normal electricity price when the sum of the maximum charging battery power and load is higher than the wind power. The grid agent regulates the DCV at V_nom via GVCM_con. The battery and wind turbine agents work in the BCCM_char and MPPT modes.

Operation NO₅: This occurs with a normal electricity price when the battery state-of-charge (SOC) level is at the maximum and the wind power is lower than the load. Also, this mode occurs with a high electricity price when the battery SOC level is at its minimum and the wind power is lower than the load. In both situations, the grid agent regulates the DCV at V_nom via GVCM_con while the battery and wind turbine agents work in the IDLE and MPPT modes.

Operation NO₆: This occurs with a normal electricity price when the battery SOC level is at its maximum and the wind power is larger than the load. Also, this mode occurs with a high electricity price when the battery SOC level is at the minimum and the wind power is higher than the load. In both cases, the grid agent regulates the DCV via GVCM_inv. The battery works in IDLE mode, and the wind turbine works in MPPT mode.

Operation NO₇: This occurs in islanded mode (IM) when the wind power is less than the sum of the load and maximum charging power of the battery. The battery agent works in BVCM (battery voltage control mode) to regulate the DCV at V_nom while the wind turbine works in MPPT mode.

Operation NO₈: This occurs in IM when the sum of the battery’s maximum charging power and load is lower than the wind power. The wind turbine works in VCM (voltage control mode) to regulate the DCV at V_nom while the battery agent operates in BCCM_char.

Operation NO₉: This occurs in IM when the battery SOC level is at its maximum, and the wind power is higher than the load. The wind turbine agent works in VCM and the battery agent operates in IDLE mode.

3.2. Control Strategy under DCV Sensor Fault

Figure 4 represents the DCV sensor fault detection and control mode decision algorithms of agent i under DCV sensor failure. In these algorithms, e_i, ${\mathsf{\epsilon }}_i$, c, c_max, ${\widehat{V}}_{DC,i}^{pre}$, $I_i^{ref}$, and $I_{i,pre}^{ref}$ represent the difference between $V_{DC,i}$ and ${\widehat{V}}_{DC,i}$, the specified threshold value, the counter value, the specified threshold of the counter, the previous value of ${\widehat{V}}_{DC,i}$, the reference current of agent i, and the previous value of $I_i^{ref}$, respectively. The variable F_fault denotes a fault flag, which is used to indicate the fault in the DCV sensor of agent $i$. When the DCV sensor has a failure, F_fault is set to one. Otherwise, it is reset to zero. Also, the variable F_mode represents a control mode flag to indicate the operation of agent i before DCV sensor fault. If the operation of agent i is in voltage control before DCV sensor fault, F_mode becomes one. Otherwise, it is reset to zero.

In the proposed strategy, when F_fault is equal to zero, the system compares $e_i$ with ${\mathsf{\epsilon }}_i$ to investigate the normal operation of the DCV sensor. When $e_i$ is larger than ${\mathsf{\epsilon }}_i$, the counter is utilized to count this event. When counter c is less than c_max, counter c increases and ${\widehat{V}}_{DC,i}$ is replaced with ${\widehat{V}}_{DC,i}^{pre}$. However, if counter c is greater than c_max, the fault detection algorithm confirms DCV sensor failure in power agent i. As soon as DCV sensor fault is identified, the fault detection algorithm sets the fault flag, F_fault, to one. Then, the control mode decision algorithm of agent i is executed to guarantee system stability even under DCV sensor fault by determining a proper operation.

In the control mode decision algorithm, transition operations to current control mode depend on the previous operations of the power agent having DCV sensor failure. If DCV sensor failure occurs in the power agent that does not regulate the DCV, it still maintains the operation of the current control mode, with the current reference being as follows:

$$I_i^{ref}=I_{i,pre}^{ref}$$

(4)

On the other hand, if DCV sensor failure happens in power agent i, which regulates the DCV, the other power agent should take the role of regulating the DCV. For this purpose, the power agent operation is shifted to current control mode with the reference current change. In case DCV sensor fault occurs in the grid or battery agent that regulates the DCV, the grid or battery agent changes its operation mode to current control mode, with the current reference being as follows:

$$I_i^{ref}=I_{i,pre}^{ref}-{\mathsf{\delta }}_i$$

(5)

Because more power is supplied to the DC bus, the DCV level is continuously increased. As a result, by detecting this DCV transition, remaining power agents can regulate the DCV after the transition operation mode. In case DCV sensor fault occurs in the wind turbine agent that regulates the DCV, the wind turbine agent changes its operation mode to current control mode, with the current reference being as follows:

$$I_W^{ref}=I_{W,pre}^{ref}+{\mathsf{\delta }}_W$$

(6)

Because the power supplied to the DC bus is reduced, the DCV level decreased. By detecting this DCV transition, remaining power agents can regulate the DCV after the transition operation mode.

To ensure both power balance and voltage regulation under DCV sensor fault,

Table 2. Additional steady-state operation modes according to DCV sensor fault.

Mode	Battery Agent	Wind Turbine Agent	Grid Agent	Load Agent
AO1	BVCM	MPPT	GCCMinv	Normal
AO2	BCCMdis	VCM	GCCMinv	Normal
AO3	BVCM	MPPT	GCCMcon	Normal
AO4	BCCMdis	VCM	GCCMcon	Normal
AO5	BCCMchar	VCM	GCCMinv	Normal
AO6	BCCMchar	VCM	GCCMcon	Normal
AO7	IDLE	VCM	GCCMcon	Normal
AO8	IDLE	VCM	GCCMinv	Normal
AO9	BCCMdis	VCM	Fault	Normal

shows additional steady-state operation modes for a decentralized DCMG system according to the DCV sensor fault. During normal operation without any failure, the DCMG operates in one of the steady-state operation modes listed in Table 1. Under DCV sensor failure, the DCMG operation is changed to one of additional the steady-state modes in Table 2 after the transition operations, which will be explained in the next subsection.

Nine additional steady-state operation modes in Table 2 are explained in detail. The operation mode transition from the normal operation in Table 1 to the additional operation in Table 2, caused by the DCV sensor faults, will be described in Figure 5. Similarly to the fact that the DC bus voltage is always regulated by at least one power agent in steady-state operation modes NO₁~NO₉ in Table 1, it is also regulated by at least one power agent in AO₁~AO₉ in Table 2.

Operation AO₁: In this mode, because DCV sensor fault occurs in the grid, the grid operates in GCCM_inv (grid current control mode by inverter operation), and the battery operates in BVCM to regulate DCV at V_nom. The wind turbine operates in MPPT mode.

Operation AO₂: In this mode, because the DCV sensor fault occurs in both the grid and battery, the battery operates in BCCM_dis and the grid operates in GCCM_inv. The wind turbine operates in VCM to regulate the DCV at V_nom.

Operation AO₃: The battery agent operates in BVCM to regulate the DCV at V_nom. The wind operates in MPPT mode and the grid agent operates in GCCM_con (grid current control mode by converter operation).

Operation AO₄: The wind turbine operates in VCM to regulate the DCV at V_nom. The battery operates in BCCM_dis and the grid operates in GCCM_con.

Operation AO₅: The wind turbine operates in VCM to regulate the DCV at V_nom. The battery operates in BCCM_char and the grid operates in GCCM_inv.

Operation AO₆: The wind turbine operates in VCM to regulate the DCV at V_nom. The battery operates in BCCM_char and the grid operates in GCCM_con.

Operation AO₇: The wind turbine operates in VCM to regulate the DCV at V_nom. The battery operates in IDLE mode and the grid operates in GCCM_con.

Operation AO₈: The wind turbine operates in VCM to regulate the DCV at V_nom. The battery operates in IDLE mode and the grid operates in GCCM_inv.

Operation AO₉: The wind turbine operates in VCM to regulate the DCV at V_nom. The battery operates in BCCM_dis.

Figure 5 shows operation mode transitions caused by DCV sensor failure, in which the DCV sensor faults of the battery, wind turbine, and grid are denoted by the variables $F_B$, $F_W$, and $F_G$, respectively. The maximum and minimum SOC levels of the battery agent are denoted by $SO{C}_B^H$ and $SO{C}_B^L$, respectively. The variables $F_B$, $F_W$, and $F_G$ are represented as follows:

$$\begin{array}{c}{F}_i=1, \mathrm{if\; DCV\; sensor\; failure\; happens\; in\; power\; agent} i\\ F_i=0, \mathrm{otherwise},\\ \mathrm{for} i=B, W, G.\end{array}$$

(7)

Similarly, two variables to denote the battery SOC status are defined as follows:

$$\begin{array}{c}{F}_{BH}=1, \mathrm{if} SO{C}_B= SO{C}_B^H;\\ F_{BH}=0, \mathrm{otherwise};\end{array}$$

(8)

$$\begin{array}{c}{F}_{BL}=1, \mathrm{if} SO{C}_B= SO{C}_B^L;\\ F_{BL}=0, \mathrm{otherwise}\end{array}$$

(9)

where $SO{C}_B$ denotes the battery SOC.

If DCV sensor failures occur during normal operation NO₁, the DCMG is changed to additional operation modes as represented in Figure 5a. During the operation of NO₁, if DCV sensor failure happens in the battery (F_B = 1) or wind turbine (F_W = 1) or both the wind turbine and battery ((F_W = 1) and (F_B = 1)), the DCMG system still maintains in NO₁ because the wind turbine and battery do not regulate the DCV before sensor fault. If the grid (F_G = 1) or both the grid and wind turbine ((F_G = 1) and (F_W = 1)) have DCV sensor fault, the DCMG system is changed to mode AO₁. The grid agent changes the control mode from GVCM_inv in NO₁ to GCCM_inv in AO₁ according to Figure 4. In this situation, the battery operates in BVCM to regulate the DCV. Finally, if both the grid and battery agents ((F_G = 1) and (F_B = 1)) have DCV sensor fault, only the wind turbine can regulate the DCV. Therefore, the DCMG system is changed to mode AO₂.

Figure 5b–d show the transition to additional operation modes of the DCMG when DCV sensor failure occurs during normal DCMG system operation in NO₂, NO₃, and NO₄, respectively. These figures show similar behavior with that shown in Figure 5a according to the location of DCV sensor failure. However, under each DCV sensor fault, the destination of additional operation modes is quite different. In the normal operation modes NO₂, NO₃, and NO₄, the grid agent regulates the DCV either in converter or inverter operation. Even in the case of (F_B = 1), (F_W = 1) or ((F_B = 1) and (F_W = 1)), the DCMG system still maintains the previous operations, NO₂, NO₃, and NO₄, respectively, as in Figure 5b–d, because the grid agent can still regulate the DCV. If DCV sensor failure happens in the grid, (F_G = 1) or ((F_G = 1) and (F_W = 1)), DCV regulation is achieved by the battery agent, which results in a transition into additional operation modes AO₃, AO₁, and AO₃, as shown in Figure 5b–d, respectively. If the failure of the DCV sensor happens in the grid and battery at the same time ((F_G = 1) and (F_B = 1)), only the wind turbine is able to regulate the DCV. Then, via a transition into additional operation modes AO₄, AO₅, and AO₆, the wind turbine regulates DCV in VCM, and other agents operate in current control mode.

Figure 5e represents the operation mode transition from NO₅ due to DCV sensor faults. As we mentioned before, NO₅ occurs with a normal electricity price when the battery SOC level is at the maximum, and the when the wind power is lower than the load. Also, this mode occurs with a high electricity price when the battery SOC level is at the minimum, and when the wind power is lower than the load. If DCV sensor failure does not occur in the grid, the DCMG maintains mode NO₅. In the condition [(F_BH = 1)&(F_G = 1)] or [(F_G = 1)&(F_B = 1)], the DCMG is changed from NO₅ to AO₇. In the first condition, the battery operates in IDLE mode because it can not absorb charging power. In the second condition, because the battery and grid can not regulate DCV, the wind turbine operates in VCM to regulate DCV, which results in AO₇. In the condition of [(F_BL = 1)&{(F_G = 1) or ((F_G = 1)&(F_W = 1))}], the DCMG system is changed from NO₅ to AO₃. This condition represents the case in which DCV sensor failure happens in the grid, or in both the grid and wind turbine agents when the battery SOC level is at the a minimum. In either case, the battery works in BVCM to regulate the DCV with charging, which results in AO₃.

Figure 5f has a similar structure to Figure 5e. While the grid agent starts with GVCM_con in the normal operation mode NO₅ in Figure 5e, it starts with GVCM_inv in NO₆ in Figure 5f. Unless DCV sensor failure occurs in the grid, the DCMG maintains NO₆. Depending on the battery’s SOC condition and the DCV sensor fault locations, this normal operation mode is shifted to AO₈ or AO₁, respectively. In AO₁, the battery agent regulates the DCV in BVCM. If the battery is not available to regulate the DCV because of the SOC condition or DCV sensor failure, the wind turbine is used in VCM, instead.

Figure 5g shows the operation mode transition caused by the DCV sensor faults when the DCMG works in NO₇. If the DCV sensor in the battery is operating normally regardless of whether DCV sensor failure occurs in the wind turbine or grid, the DCMG maintains NO₇. However, when the battery has a DCV sensor failure, i.e., (F_B = 1), only the wind turbine can regulate the DCV in VCM. Then, the DCMG system is changed to AO₉ or NO₈ depending on the relation of the wind power and load.

Figure 5h represents transition of the DCMG system when DCV sensor failure occurs during the normal operation NO₈. Unless DCV sensor failure occurs in the wind turbine, the DCMG maintains NO₈. If the wind turbine has a DCV sensor failure (F_W = 1), the battery agent regulates the DCV in BVCM, which results in the change of the DCMG operation to NO₇.

Finally, Figure 5i represents the operation transition from NO₉ caused by DCV sensor failure. Since only the wind turbine is operating in NO₉, the DCMG maintains NO₉, if the DCV sensor of the wind turbine does not have a fault.

3.3. Transitional Operation Modes in Power Agents

In the proposed method, all power agents in the DCMG determine the operation modes by detecting the information of the DCV level without DCLs. When certain transient conditions are introduced into the DCMG, the agents of the DCMG system temporarily shifts the DCV value to different levels during a predetermined period. During this predetermined time, all the power agents appropriately change the operation mode during the transient periods before the DCMG system returns to another steady-state operation mode.

Table 3. Power agents to activate each transition operation.

Mode	Grid Agent	Wind Turbine Agent	Battery Agent	Agent Action
TO1	✔			▪ Grid agent activates TO1. ▪ Grid agent regulates DCV to VL1. ▪ If DCV level reaches VL1, grid agent keeps DCV at VL1 for 0.5 s. ▪ As DCV level reaches VL1, and lasts for 0.1 s, other power agents identify this event, and change their operations.
TO2	✔			▪ Grid agent activates TO2. ▪ Grid agent regulates DCV to VH1. ▪ If DCV level reaches VH1, grid agent keeps DCV at VH1 for 0.6 s. ▪ As DCV level reaches VH1, and lasts for 0.1 s, other power agents identify this event, and change their operations.
TO3	✔		✔	▪ Grid or battery agent activates TO3. ▪ DCV increases due to surplus power. ▪ If DCV level reaches VH3, wind turbine agent keeps DCV at VH3 for 0.3 s. ▪ As DCV level reaches VH3, and lasts for 0.01 s, battery agent identifies this event, and changes its operations.
TO4	✔	✔		▪ Grid or wind turbine agent activates TO4. ▪ DCV decreases due to deficient power. ▪ If battery agent detects that DCV level is kept lower than VL2 for more than 0.01 s, battery operation is changed to BVCM.
TO5	✔			▪ Grid agent activates TO5. ▪ DCV increases due to surplus power. ▪ If battery agent detects that DCV level is kept higher than VH2 for more than 0.01 s, battery operation is changed to BVCM.
TO6	✔	✔	✔	▪ All power agents activate TO6. ▪ DCV decreases due to deficient power. ▪ If load agent detects that DCV level is kept lower than VL3 for more than 0.01 s, load shedding mode is activated.

lists the transition operation modes that are introduced into the DCMG system due to DCMG system uncertainty. This table also shows the power agents that activate each transition operation mode and resultant actions. In this table, V_L1, V_L2, and V_L3 denote the first, second, and third low-level DCV, respectively, and V_H1, V_H2, and V_H3 denote the first, second, and third high-level DCV, respectively.

When an event occurs such as the electricity price changing from normal to high, or grid agent reconnection from a fault with a high electricity price happens, the DCMG system temporarily uses the transition operation mode TO₁. In TO₁, the grid agent regulates the DCV to V_L1 to inform other power agents of this event. Figure 6 shows the detection of the transition operation modes by the power agent. In Figure 6, the horizontal axis represents the time required for transition operation detection, the vertical axis represents the DCV levels for different transition operations, and the color indicates the power agents that detect the transition operation. As the DCV level reaches V_L1, and lasts for 0.1 s, the other power agents such as the battery, wind turbine, and load agent identify this event, and change their operation appropriately. After 0.5 s from the instant that the DCV reaches V_L1, the grid regulates the DCV back to V_nom.

When an event occurs such as a change in the electricity price from high to normal, or when the reconnection of the grid agent from a fault with a normal electricity price happens, the DCMG system temporarily uses the transition operation mode TO₂. In TO₂, the grid regulates the DCV to V_H1 to inform other power agents of this event. As shown in Figure 6, as the DCV level reaches V_H1, and lasts for 0.1 s, the other power agents identify this event, and change their operation appropriately. After 0.6 s from the instant that the DCV reaches V_H1, the grid agent regulates the DCV back to V_nom.

In the IM of the DCMG system, if the battery is in critical condition, such as when it has a maximum battery SOC_B, the DCV increases since the battery cannot absorb power. After the DCV reaches V_H3, the wind turbine agent maintains the DCV at V_H3 for 0.3 s with the transition operation mode TO₃. After 0.01 s from the instant that the DCV reaches V_H3, the battery agent identifies this event to change the operation appropriately. After 0.3 s, the wind turbine agent regulates the DCV to V_nom again. This transition operation mode, TO₃, is also initiated by the grid when the grid operating in GVCM_inv is disconnected from the DCMG system, or when the grid agent has a DCV sensor fault.

The transition operation mode TO₄ is triggered by the grid when the grid operating in GVCM_con has a fault. The transition operation mode TO₄ is also activated when the wind power is decreased in IM. In this situation, since the supply power is smaller than the load, the DCV decreases. If the battery agent detects that the DCV level is kept lower than V_L2 for more than 0.01 s, as shown in Table 3, the battery agent acknowledges this event. Then, the battery changes operation to BVCM to regulate the DCV back to V_nom.

When the grid agent is in emergency conditions such as a grid fault under a high electricity price or DCV sensor fault, the DCV increases due to surplus power, which triggers the transition operation mode TO₅. As the DCV level is kept higher than V_H2 for more than 0.01 s, the battery agent acknowledges this event, and changes operation to BVCM to regulate the DCV back to V_nom.

When SOC_B reaches the minimum level and the wind generation power is smaller than the load demand in the islanded DCMG, the DCV is decreased to the critical value of V_L3. After 0.01 s from the instant the DCV reaches V_L3, load shedding starts.

4. Simulation Tests

In this section, the reliability and feasibility of the proposed strategy are evaluated through simulation tests by using PSIM tools. The simulations are conducted considering several uncertain operating conditions such as DCV sensor fault, grid fault, minimum or maximum battery SOC, agent power variation, and electricity price change in the grid. System parameters of a decentralized DCMG are shown in

Table 4. DCMG parameters.

Power Agents	Parameters	Values
Battery	$\mathrm{Minimum} \mathrm{SOC} \left(SO{C}_B^L\right)$	20%
	$\mathrm{Maximum} \mathrm{SOC} \left(SO{C}_B^H\right)$	90%
	Maximum discharging power	−540 W
	Maximum charging power	540 W
	Maximum voltage	180 V
	Rated capacity	25 Ah
	Converter filter inductance, L	7 mH
Grid	Transformer, Y/Δ	380/220 V
	Grid frequency	60 Hz
	Grid voltage	220 V
	LCL filter	1.7 mH/4.5 μF/1.7 mH
Wind turbine	PMSG number of poles	6
	PMSG inertia	0.111 kgm2
	PMSG stator resistance	0.64 Ω
	PMSG dq-axis inductance	0.82 mH
	PMSG flux linkage	0.18 Wb
	Converter filter inductance	7 mH
Load	Load 1	200 W
	Load 2	200 W
	Load 3	200 W
	Priority: load 1 > load 2 > load 3	-
DC bus	Nominal DCV (Vnom)	400 V
	First level of high DCV (VH1)	405 V
	Second level of high DCV (VH2)	410 V
	Third level of high DCV (VH3)	415 V
	First level of low DCV (VL1)	390 V
	Second level of low DCV (VL2)	380 V
	Third level of low DCV (VL3)	370 V
	Capacitance	4 mF

. In this table, PMSG denotes a permanent magnet synchronous generator, and the LCL filter represents the filter composed of an inductor, capacitor, and inductor.

4.1. Transition from Grid-Connected Mode to IM under DCV Sensor Fault

Figure 7 represents simulation test results for when the DCMG undergoes a transition from grid-connected mode to IM, and the battery has DCV sensor failure. In Figure 7, ${\widehat{V}}_{DC,i}$ (for i = G, W, and B) is the estimated DCV value in the power agent, $V_{DC,i}^{sensor}$ is the DCV measured by the sensor in the power agent, and $V_{DC}^{real}$ is the actual DCV value. In this simulation test, the DCMG starts in NO₄, which is one of the GCMs. In this mode, the grid operates in GVCM_con to regulate the DCV at V_nom, the battery operates in BCCM_char, and the wind turbine agent operates in MPPT mode. At t = 0.5 s, grid fault occurs, which causes the change of the DCMG system operation from GCM NO₄ to IM TO₄. Since the supplied power is smaller than the load, the DCV is reduced. If the battery agent detects that the DCV level is kept lower than V_L2 for more than 0.01 s, as shown in Table 3, the battery acknowledges grid disconnection from the DCMG. Then, the battery changes operation from BCCM_char to BVCM to regulate the DCV back to V_nom, and the entire DCMG system operation is changed from TO₄ to NO₇.

At t = 1.2 s, when the DCV is regulated by the battery, battery DCV sensor fault occurs. As soon as the battery agent identifies DCV sensor failure, it is shifted to current control mode by the proposed fault detection and control mode decision algorithms shown in Figure 4. By using the proposed strategy, the battery agent increases the DCV by supplying more power to the DC bus. If the DCV level reaches V_H3, the wind turbine agent maintains the DCV to V_H3 for 0.3 s. Eventually, the DCMG is changed from NO₇ to TO₃. After 0.3 s from the instant that the DCV reaches V_H3, the wind turbine agent regulates the DCV back to V_nom and subsequently the DCMG system operation is changed from TO₃ to AO₉.

This simulation test confirms that the battery agent can still supply power to the DC bus in current control mode even if the battery has a DCV sensor failure, which clearly demonstrates that the proposed control strategy ensures power sharing and seamless power management even in abnormal system conditions.

4.2. Transition from Normal to High Electricity Price Conditions under DCV Sensor Fault

Figure 8 represents simulation test results for when the electricity price changes from normal to high under grid DCV sensor failure. The DCMG initially works in NO₄ with a normal electricity price. In this mode, the grid operates in GVCM_con to regulate the DCV at V_nom, the battery operates in BCCM_char, and the wind turbine operates in MPPT mode. At t = 0.3 s, when the electricity price is changed from normal to high, the grid agent regulates the DCV to V_L1 in order for the other agents to recognize the electricity price change. Then, the DCMG is changed from NO₄ to TO₁. After 0.1 s from the instant that the DCV reaches V_L1, the wind turbine and battery detect this event. Then, the battery changes operation from BCCM_char to BCCM_dis to inject power to the DC bus, which causes the grid to change from GVCM_con to GVCM_inv. After 0.5 s from the instant that the DCV reaches V_L1, the grid regulates the DCV back to V_nom, and the entire DCMG operation is changed from TO₁ to NO₁.

At t = 1.5 s, when grid DCV sensor fault occurs, the grid operation is changed to current control mode as soon as DCV sensor failure is detected, as shown in Figure 4. After that, the DCMG system is changed from NO₁ to TO₅. According to the proposed fault detection and control mode decision algorithms, the grid agent increases the DCV by injecting more power to the DC bus. When the battery detects that the DCV level is kept higher than V_H2 for more than 0.01 s, the battery recognizes the DCV sensor fault in the grid. Then, the battery changes operation from BCCM_dis to BVCM to regulate the DCV back to V_nom, and the DCMG is changed from TO₅ to AO₁. This simulation test result also validates continuous power sharing in the presence of failure in the grid DCV sensor.

4.3. Grid Recovery with High Electricity Price Condition under DCV Sensor Fault

Figure 9 represents the simulation test results in the case of grid reconnection at a high electricity price under wind DCV sensor failure. The DCMG is assumed to start in NO₇, which is one of the IMs. In this mode, the battery works in BVCM to regulate the DCV at V_nom, and the wind turbine operates in MPPT mode. At t = 0.5 s, DCV sensor failure occurs in the wind turbine. However, the DCMG system maintains mode NO₇ since the wind turbine does not regulate the DCV. In this circumstance, the wind turbine continues operating in MPPT mode with $I_W^{ref}=I_{W,pre}^{ref}$, as explained in Figure 4.

At t = 1.0 s, when the grid agent reconnects at a high electricity price, the grid works in GVCM_inv to regulate the DCV to V_L1, and the DCMG is changed from NO₇ to TO₁. After 0.1 s from the instant that the DCV reaches V_L1, the wind turbine and battery detect grid reconnection at a high electricity price. Then, the battery changes operation from BVCM to BCCM_dis to supply the power to the DC bus. After 0.5 s from the instant that the DCV reaches V_L1, the grid regulates the DCV back to V_nom, and the entire DCMG operation is changed from TO₁ to NO₁. This simulation result also proves the occurrence of continuous power sharing under DCV sensor failure in the wind turbine.

4.4. Transition of Electricity Price Condition from High to Normal under DCV Sensor Fault

Figure 10 represents simulation test results for when the electricity price changes from high to normal under battery and grid DCV sensor faults. In this simulation, the DCMG initially operates at NO₁ with a high electricity price. The grid agent operates in GVCM_inv to regulate the DCV at V_nom, the battery agent operates in BCCM_dis, and the wind turbine operates in MPPT mode.

At t = 0.3 s, when the electricity price changes from high to normal, the grid agent regulates the DCV to V_H1, and the DCMG system is changed from NO₁ to TO₂. After 0.1 s from the instant that the DCV reaches V_H1, the wind turbine and battery detect the change in electricity price. Then, the battery agent changes operation from BCCM_dis to BCCM_char to absorb power from the DC bus, which causes the grid to change operation from GVCM_inv to GVCM_con. After 0.6 s from that instant the DCV reaches V_H1, the grid regulates the DCV back to V_nom, and the DCMG is changed from TO₂ to NO₄.

At t = 1.4 s, DCV sensor failure occurs in the battery. The DCMG maintains NO₄ because the battery does not regulate DCV. In this case, the battery agent maintains the previous operation BCCM_char with $I_B^{ref}=I_{B,pre}^{ref}$ as shown in Figure 4.

At t = 1.7 s, DCV sensor failure happens in the grid that is regulating the DCV to V_nom. When the grid agent detects DCV sensor failure, the grid operation is instantly shifted to current control mode by the proposed scheme, which changes the DCMG operation from NO₄ to TO₃. The grid agent increases the DCV by injecting larger power into the DC bus. If the battery agent detects that the DCV level is kept higher than V_H2 for more than 0.01 s, the battery agent acknowledges the grid DCV sensor fault. However, the battery cannot regulate the DCV due to DCV sensor failure. In this case, the DCV level further increases to more than V_H2. As the DCV reaches V_H3, the wind turbine instantly maintains the DCV at V_H3 for 0.3 s. After 0.3 s from the instant that the DCV reaches V_H3, the wind turbine regulates the DCV back to V_nom, and the entire DCMG operation is changed from TO₃ to AO₆. This simulation test shows that the proposed scheme ensures uninterruptible power management even in circumstances of multiple failures in a DCV sensor.

4.5. Case of Minimum Battery SOC Level

Figure 11 represents simulation test results for load shedding when the battery SOC level reaches the minimum. The DCMG starts in NO₇, which is one of the IMs. In this mode, the wind power is smaller than the load, which causes the battery to operate in BVCM by discharging to maintain the DCV at V_nom.

At t = 0.6 s, when the battery SOC level drops to the minimum value, the battery changes operation from BVCM to IDLE mode, and the DCMG is changed from NO₇ to TO₆. Since the power supply is smaller than the load, the DCV drops rapidly. If the load agent detects that the DCV level is kept lower than V_L3 for more than 0.01 s, as shown in Table 3, the load agent recognizes that the supplied power is less than the load. Consequently, load shedding is activated to prevent DCMG system collapse in these emergency conditions. When the DCV becomes higher than V_nom because of load shedding, the battery operation is changed from IDLE mode to BVCM to regulate the DCV back to V_nom, which changes the DCMG operation from TO₆ to NO₇.

At t = 1.3 s, as the grid agent is reconnected to the DCMG system with a normal electricity price, the grid operates in GVCM_con to regulate the DCV to V_H1, and the DCMG operation is changed from NO₇ to TO₂. After 0.1 s from the instant that the DCV reaches V_H1, the wind turbine, load, and battery recognize grid reconnection with a normal electricity price. Consequently, the battery changes operation from BVCM to BCCM_char and load reconnection starts. After 0.6 s from the instant that the DCV reaches V_H1, the grid regulates the DCV back to V_nom, and the entire DCMG operation is changed from TO₂ to NO₄.

5. Experimental Results

An experimental hardware setup of the decentralized DCMG that is depicted in Figure 12 is utilized to confirm the feasibility and reliability of the proposed strategy. To construct a decentralized DCMG, four power agents, which are the wind turbine, grid, battery, and load agents, are used with the system parameters in Table 4. The wind turbine agent is constructed with a unidirectional AC–DC converter and wind turbine emulator that is composed of a PMSG, a mechanically coupled induction motor, and an AC motor control panel. The battery agent employs an interleaved bidirectional DC–DC converter to connect the battery with the DC bus. The grid agent is composed of a bidirectional AC–DC converter, main grid, and a Y-Δ transformer. A digital signal processor, TMS320F28335, is used to realize the proposed strategy. The experimental tests are conducted under five different conditions: change in electricity price from normal to high, DCV sensor failure in GCM, transition from grid-connected mode to IM, minimum SOC_B level in IM, and DCV sensor fault in IM.

5.1. Transition of Electricity Price from Normal to High in GCM

Figure 13 represents the experimental tests in the case that the electricity price changes from high to normal in GCM. The DCMG system starts in NO₅, which is one of the GCMs, and SOC_B is at the maximum level. The grid operates in GVCM_con to regulate the DCV to V_nom, the wind turbine works in MPPT mode, and the battery works in IDLE mode.

When the electricity price changes from normal to high, the grid regulates the DCV to V_L1 in order for the other agents to recognize the change in electricity price. Consequently, the DCMG operation is changed from NO₅ to TO₁. After 0.1 s from the instant that the DCV reaches V_L1, the wind turbine and battery detect the change in electricity. Then, the battery changes the operation from IDLE mode to BCCM_dis to supply power to the DC bus, which makes the grid change operation from GVCM_con to GVCM_inv. After 0.5 s from the instant the DCV reaches V_L1, the grid regulates the DCV back to V_nom, and the entire DCMG operation is changed from TO₁ to NO₁. This experimental result clearly shows voltage stabilization and good power balance using the decentralized control method without any DCLs.

5.2. GCM under DCV Sensor Fault

Figure 14 represents the experimental test for GCM under grid and battery DCV sensor faults. In this test, the DCMG system initially operates in NO₄ with a normal electricity price. In this mode, the grid operates in GVCM_con to regulate the DCV at V_nom, while the battery and wind turbine operate in BCCM_char and MPPT mode.

When a DCV sensor fault happens in a battery, the DCMG system stays in NO₄ because the battery does not regulate the DCV. In this situation, the battery operation is still BCCM_char with $I_B^{ref}=I_{B,pre}^{ref}$, as shown in Figure 4. When the grid agent that is regulating the DC bus has a sensor failure, the grid operation mode is changed to GCCM_con, and the DCMG system operation is changed from NO₄ to TO₃. Similar to what is shown in Figure 10, the DCV increases rapidly in this case since the grid supplies larger power to the DC bus. If the battery agent detects that the DCV level is kept higher than V_H2 for more than 0.01 s, the battery agent acknowledges grid DCV sensor fault. However, the battery cannot regulate the DCV due to DCV sensor failure. In this case, the DCV level further increases to more than V_H2. As the DCV reaches V_H3, the wind turbine instantly operates in VCM to maintain the DCV at V_H3. After 0.3 s from the instant that the DCV reaches V_H3, the wind turbine regulates the DCV back to V_nom, and the DCMG operation is changed from TO₃ to AO₆. This experimental test shows the same behaviors as those in the simulation in Figure 10, which verifies that the proposed control strategy achieves both voltage stabilization and power balance even under multiple failures in DCV sensors.

5.3. Transition between GCM and IM

Figure 15 shows the experimental test for the transition from GCM to IM. The DCMG operates in NO₄, which is one of the GCMs. The grid operates in GVCM_con to regulate the DCV to V_nom, while the battery and wind turbine operate in BCCM_char and MPPT mode.

When the grid is disconnected from DCMG, it causes a change of DCMG operations from GCM NO₄ to IM TO₄. The DCV decreases rapidly since the power supply is smaller than the power demand. If the battery detects that the DCV level is kept lower than V_L2 for more than 0.01 s, the battery acknowledges grid disconnection from the DCMG system. Then, the battery changes operation from BCCM_char to BVCM to regulate the DCV back to V_nom, and the entire DCMG system operation is changed from TO₄ to NO₇.

5.4. Test with IM under Minimum Battery SOC Level

Figure 16 shows the experimental test for load shedding when the battery SOC level reaches the minimum value. The DCMG starts in NO₇, which is one of the IMs. In this mode, the battery supplies power to the DC bus to regulate the DCV in BVCM since the wind power is smaller than the load. When the battery SOC level reaches the minimum value, the battery changes the operation from BVCM to IDLE mode, and the DCMG is changed from NO₇ to TO₆. The DCV is decreased rapidly because the power supply is smaller than the power demand. After 0.01 s from the instant that the DCV is lower than V_L3, the load agent detects that the power supply is smaller than the demand, and then load shedding is activated to prevent DCMG system collapse for such an emergency condition. When the DCV becomes higher than V_nom, because the power supply is higher than the demand, the battery operation is changed from IDLE mode to BVCM to regulate the DCV back to V_nom, which changes the DCMG system from TO₆ to NO₇.

5.5. Test with IM under DCV Sensor Fault

Figure 17 shows the experimental test for the IM under battery DCV sensor failure. Initially, the DCMG system starts in NO₇, which is one of the IMs. In this mode, the battery supplies power to the DC bus to regulate the DCV in BVCM since the wind power is smaller than the load. When battery DCV sensor fault occurs, the battery identifies DCV sensor failure. Then, the battery changes the operation to current control mode. By using the proposed scheme, the battery increases the DCV, and the DCMG is changed from NO₇ to TO₃. When the DCV reaches V_H3, the wind turbine regulates the DCV to V_H3 in VCM. After 0.3 s from the instant that the DCV reaches V_H3, the wind turbine regulates the DCV back to V_nom, and the DCMG operation is changed from TO₃ to AO₉.

6. Conclusions

This study proposed a continuous power management strategy for a decentralized DCMG based on transitional operation modes under system uncertainty and sensor failure. The power agents connected in the decentralized DCMG utilize only a primary controller to achieve both voltage regulation and power sharing with reduced computational burden. To improve the reliability of DCMG operation, the abnormality of the DCV sensor is monitored by the proposed DCV sensor fault detection algorithm, in which all the power agents estimate the DCV by using the observer and compare the estimated DCV with the measured DCV. If a power agent detects the failure of the DCV sensor, it modifies the operational mode appropriately to ensure the stability of the DCMG in accordance with the proposed control mode decision algorithm under the DCV sensor fault. The proposed algorithms can maintain the operation of the DCMG system without system collapse even with multiple sensor failures, if only one sensor operates normally.

To achieve continuous power balance as well as voltage stabilization even under several uncertainty conditions such as power variation in distributed generation, sudden grid disconnection, and critical battery status, the DCMG system uses the transitional operation modes to transmit the proper information to other power agents without using DCLs in this study. In the transitional operation modes, the DCV levels are temporarily shifted to an appropriate level during the predetermined period. Based on the DCV levels, each power agent determines its operation mode after detecting the uncertain conditions. The effectiveness and reliability of the proposed control strategy were demonstrated via various simulation and experimental tests under several test conditions. In summary, the experimental tests clearly demonstrated that DCV regulation could be achieved rapidly even under uncertain conditions. In the experimental tests, when multiple DCV sensor faults occurred, the transition time to restore the DCV to its nominal value was 1.6 s under the proposed strategy. In the grid fault condition, the DCV was restored in 1.0 s. In addition, when the battery reached the minimum SOC level, the transition time to restore the DCV to its nominal value was 1.8 s. In the case of DCV sensor failure in islanded mode, the transition time was 1.9 s.